access2rohit opened a new issue #18947: URL: https://github.com/apache/incubator-mxnet/issues/18947
## Description Use int64_t for data index type for internal MXNet operators and ensure 3rd party dependencies like BLAS/LAPACK libraries, OpenCV, oneDNN(oneAPI) and cuDNN also accept `long long/int64_t` parameters for index types. it should be a 0 change for end user (training/inference)scripts. Tasks [x] Investigate and build openBLAS with int64 support [] Investigate building MXNet with MKL as BLAS with ILP64 support(Engage Intel to contribute) [] Investigate openCV int64 support [] cuDNN to support large inputs(ask Nvidia) [] Support all ops under np namespace to work with Large inputs [] Support all ops under npx namespace to work with Large inputs [] Enable int64 versions on BLAS engine and openCV to work with Large tensor enabled MXNet in CI as default [] Add nightly tests for npx operators only (these are core deeplearning ops) ## Team @Zha0q1 @josephevans @access2rohit ---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected]
